,TABLE OF CONTENTS
1. Introduction........................................................................................................................................................................... 2
2. Descriptive statistics.......................................................................................................................................................... 2
3. Logistic regression model................................................................................................................................................ 8
3.1 Model..................................................................................................................................................................................... 8
3.2 Results................................................................................................................................................................................... 9
3.2 Validation.......................................................................................................................................................................... 11
4. Classification trees............................................................................................................................................................ 12
4.1 CRT....................................................................................................................................................................................... 12
4.2 CHAID.................................................................................................................................................................................. 14
4.3 Validation.......................................................................................................................................................................... 17
5.Recommendation and critical reflection................................................................................... 17
REFERENCES........................................................................................................................................... 20
1
, 1. Introduction
The Orange County Register (OC Register) is a newspaper that covers the local news
in Orange County, California, United States. While they have a healthy customer base of
about 250,000 subscribers, they face a declining trend in circulation, as well as many other
newspaper companies. An analysis from PEW Research Center shows that the total weekday
circulation for U.S. daily newspapers fell 8% in 2016, marking the 28 th consecutive year of
declines (Barthel, 2017).
In order to manage the declining trend in circulation, OC Register is engaging actively
to maintain and expand its current customer base. They already focused on creating and
maintaining online channels by introducing an E-edition of the newspaper for digital devices
and constantly updating their extensive website. Additionally, OC Register’s marketing
department wants to emphasize retention of customers even more. At the moment they
actively try to persuade customers not to cancel their subscription – i.e. they offer discounts
on the digital edition to physical subscribers who might cancel their subscription. Whereas
price discounts might be effective, it is proven to reduce profitability. That is why OC
Register wants to identify which customers are most likely to churn. After identification, the
newspaper wants to initiate a discount campaign targeted at customers with a high likelihood
of switching in order to retain them.
Therefore, the purpose of this report is to identify which customers should be targeted
for this campaign by predicting the churn behavior of a sample of customers, while
simultaneously selecting the best model to predict churn, so that it can be used at a later point
in time when a similar marketing initiative is run again.
2. Descriptive statistics
Before building the models to predict churn behavior, the data provided by OC
Register needs to be assessed. The dataset is collected in 2016 and contains 15.855
observations at the individual customer level. This implies that each observation represents
the behavior of one customer related to a specific variable. This level of observation is useful
to measure for example individual spending, purchase probability and in this case, the
probability to churn.
2
1. Introduction........................................................................................................................................................................... 2
2. Descriptive statistics.......................................................................................................................................................... 2
3. Logistic regression model................................................................................................................................................ 8
3.1 Model..................................................................................................................................................................................... 8
3.2 Results................................................................................................................................................................................... 9
3.2 Validation.......................................................................................................................................................................... 11
4. Classification trees............................................................................................................................................................ 12
4.1 CRT....................................................................................................................................................................................... 12
4.2 CHAID.................................................................................................................................................................................. 14
4.3 Validation.......................................................................................................................................................................... 17
5.Recommendation and critical reflection................................................................................... 17
REFERENCES........................................................................................................................................... 20
1
, 1. Introduction
The Orange County Register (OC Register) is a newspaper that covers the local news
in Orange County, California, United States. While they have a healthy customer base of
about 250,000 subscribers, they face a declining trend in circulation, as well as many other
newspaper companies. An analysis from PEW Research Center shows that the total weekday
circulation for U.S. daily newspapers fell 8% in 2016, marking the 28 th consecutive year of
declines (Barthel, 2017).
In order to manage the declining trend in circulation, OC Register is engaging actively
to maintain and expand its current customer base. They already focused on creating and
maintaining online channels by introducing an E-edition of the newspaper for digital devices
and constantly updating their extensive website. Additionally, OC Register’s marketing
department wants to emphasize retention of customers even more. At the moment they
actively try to persuade customers not to cancel their subscription – i.e. they offer discounts
on the digital edition to physical subscribers who might cancel their subscription. Whereas
price discounts might be effective, it is proven to reduce profitability. That is why OC
Register wants to identify which customers are most likely to churn. After identification, the
newspaper wants to initiate a discount campaign targeted at customers with a high likelihood
of switching in order to retain them.
Therefore, the purpose of this report is to identify which customers should be targeted
for this campaign by predicting the churn behavior of a sample of customers, while
simultaneously selecting the best model to predict churn, so that it can be used at a later point
in time when a similar marketing initiative is run again.
2. Descriptive statistics
Before building the models to predict churn behavior, the data provided by OC
Register needs to be assessed. The dataset is collected in 2016 and contains 15.855
observations at the individual customer level. This implies that each observation represents
the behavior of one customer related to a specific variable. This level of observation is useful
to measure for example individual spending, purchase probability and in this case, the
probability to churn.
2
